avcodec/x86/vp8dsp: Remove obsolete MMX(EXT) functions

The only systems which benefit from these are truely
ancient 32bit x86s as all other systems use at least the SSE2 versions
(this includes all x64 cpus (which is why this code is restricted
to x86-32)).

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

3 files changed, 4 insertions, 553 deletions

diff --git a/libavcodec/x86/vp8dsp.asm b/libavcodec/x86/vp8dsp.asm
index 75de5690a1..1c59e884ed 100644
--- a/libavcodec/x86/vp8dsp.asm
+++ b/libavcodec/x86/vp8dsp.asm
@@ -840,25 +840,6 @@ cglobal put_vp8_pixels8, 5, 5, 0, dst, dststride, src, srcstride, height
     jg .nextrow
     REP_RET
 
-%if ARCH_X86_32
-INIT_MMX mmx
-cglobal put_vp8_pixels16, 5, 5, 0, dst, dststride, src, srcstride, height
-.nextrow:
-    movq    mm0, [srcq+srcstrideq*0+0]
-    movq    mm1, [srcq+srcstrideq*0+8]
-    movq    mm2, [srcq+srcstrideq*1+0]
-    movq    mm3, [srcq+srcstrideq*1+8]
-    lea    srcq, [srcq+srcstrideq*2]
-    movq [dstq+dststrideq*0+0], mm0
-    movq [dstq+dststrideq*0+8], mm1
-    movq [dstq+dststrideq*1+0], mm2
-    movq [dstq+dststrideq*1+8], mm3
-    lea    dstq, [dstq+dststrideq*2]
-    sub heightd, 2
-    jg .nextrow
-    REP_RET
-%endif
-
 INIT_XMM sse
 cglobal put_vp8_pixels16, 5, 5, 2, dst, dststride, src, srcstride, height
 .nextrow:
@@ -895,32 +876,6 @@ cglobal put_vp8_pixels16, 5, 5, 2, dst, dststride, src, srcstride, height
     %4 [dst2q+strideq+%3], m5
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-cglobal vp8_idct_dc_add, 3, 3, 0, dst, block, stride
-    ; load data
-    movd       m0, [blockq]
-
-    ; calculate DC
-    paddw      m0, [pw_4]
-    pxor       m1, m1
-    psraw      m0, 3
-    movd [blockq], m1
-    psubw      m1, m0
-    packuswb   m0, m0
-    packuswb   m1, m1
-    punpcklbw  m0, m0
-    punpcklbw  m1, m1
-    punpcklwd  m0, m0
-    punpcklwd  m1, m1
-
-    ; add DC
-    DEFINE_ARGS dst1, dst2, stride
-    lea     dst2q, [dst1q+strideq*2]
-    ADD_DC     m0, m1, 0, movh
-    RET
-%endif
-
 %macro VP8_IDCT_DC_ADD 0
 cglobal vp8_idct_dc_add, 3, 3, 6, dst, block, stride
     ; load data
@@ -971,44 +926,6 @@ VP8_IDCT_DC_ADD
 ; void ff_vp8_idct_dc_add4y_<opt>(uint8_t *dst, int16_t block[4][16], ptrdiff_t stride);
 ;-----------------------------------------------------------------------------
 
-%if ARCH_X86_32
-INIT_MMX mmx
-cglobal vp8_idct_dc_add4y, 3, 3, 0, dst, block, stride
-    ; load data
-    movd      m0, [blockq+32*0] ; A
-    movd      m1, [blockq+32*2] ; C
-    punpcklwd m0, [blockq+32*1] ; A B
-    punpcklwd m1, [blockq+32*3] ; C D
-    punpckldq m0, m1        ; A B C D
-    pxor      m6, m6
-
-    ; calculate DC
-    paddw     m0, [pw_4]
-    movd [blockq+32*0], m6
-    movd [blockq+32*1], m6
-    movd [blockq+32*2], m6
-    movd [blockq+32*3], m6
-    psraw     m0, 3
-    psubw     m6, m0
-    packuswb  m0, m0
-    packuswb  m6, m6
-    punpcklbw m0, m0 ; AABBCCDD
-    punpcklbw m6, m6 ; AABBCCDD
-    movq      m1, m0
-    movq      m7, m6
-    punpcklbw m0, m0 ; AAAABBBB
-    punpckhbw m1, m1 ; CCCCDDDD
-    punpcklbw m6, m6 ; AAAABBBB
-    punpckhbw m7, m7 ; CCCCDDDD
-
-    ; add DC
-    DEFINE_ARGS dst1, dst2, stride
-    lea    dst2q, [dst1q+strideq*2]
-    ADD_DC    m0, m6, 0, mova
-    ADD_DC    m1, m7, 8, mova
-    RET
-%endif
-
 INIT_XMM sse2
 cglobal vp8_idct_dc_add4y, 3, 3, 6, dst, block, stride
     ; load data
@@ -1117,7 +1034,7 @@ cglobal vp8_idct_dc_add4uv, 3, 3, 0, dst, block, stride
     SWAP                 %4,  %3
 %endmacro
 
-%macro VP8_IDCT_ADD 0
+INIT_MMX sse
 cglobal vp8_idct_add, 3, 3, 0, dst, block, stride
     ; load block data
     movq         m0, [blockq+ 0]
@@ -1126,17 +1043,9 @@ cglobal vp8_idct_add, 3, 3, 0, dst, block, stride
     movq         m3, [blockq+24]
     movq         m6, [pw_20091]
     movq         m7, [pw_17734]
-%if cpuflag(sse)
     xorps      xmm0, xmm0
     movaps [blockq+ 0], xmm0
     movaps [blockq+16], xmm0
-%else
-    pxor         m4, m4
-    movq [blockq+ 0], m4
-    movq [blockq+ 8], m4
-    movq [blockq+16], m4
-    movq [blockq+24], m4
-%endif
 
     ; actual IDCT
     VP8_IDCT_TRANSFORM4x4_1D 0, 1, 2, 3, 4, 5
@@ -1153,14 +1062,6 @@ cglobal vp8_idct_add, 3, 3, 0, dst, block, stride
     STORE_DIFFx2 m2, m3, m6, m7, m4, 3, dst2q, strideq
 
     RET
-%endmacro
-
-%if ARCH_X86_32
-INIT_MMX mmx
-VP8_IDCT_ADD
-%endif
-INIT_MMX sse
-VP8_IDCT_ADD
 
 ;-----------------------------------------------------------------------------
 ; void ff_vp8_luma_dc_wht(int16_t block[4][4][16], int16_t dc[16])
@@ -1193,23 +1094,15 @@ VP8_IDCT_ADD
     SWAP %1, %4, %3
 %endmacro
 
-%macro VP8_DC_WHT 0
+INIT_MMX sse
 cglobal vp8_luma_dc_wht, 2, 3, 0, block, dc1, dc2
     movq          m0, [dc1q]
     movq          m1, [dc1q+8]
     movq          m2, [dc1q+16]
     movq          m3, [dc1q+24]
-%if cpuflag(sse)
     xorps      xmm0, xmm0
     movaps [dc1q+ 0], xmm0
     movaps [dc1q+16], xmm0
-%else
-    pxor         m4, m4
-    movq  [dc1q+ 0], m4
-    movq  [dc1q+ 8], m4
-    movq  [dc1q+16], m4
-    movq  [dc1q+24], m4
-%endif
     HADAMARD4_1D  0, 1, 2, 3
     TRANSPOSE4x4W 0, 1, 2, 3, 4
     paddw         m0, [pw_3]
@@ -1221,11 +1114,3 @@ cglobal vp8_luma_dc_wht, 2, 3, 0, block, dc1, dc2
     SCATTER_WHT   0, 1, 0
     SCATTER_WHT   2, 3, 2
     RET
-%endmacro
-
-%if ARCH_X86_32
-INIT_MMX mmx
-VP8_DC_WHT
-%endif
-INIT_MMX sse
-VP8_DC_WHT
diff --git a/libavcodec/x86/vp8dsp_init.c b/libavcodec/x86/vp8dsp_init.c
index 0238898e58..cceb346ced 100644
--- a/libavcodec/x86/vp8dsp_init.c
+++ b/libavcodec/x86/vp8dsp_init.c
@@ -112,9 +112,6 @@ void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, ptrdiff_t dststride,
 void ff_put_vp8_pixels8_mmx (uint8_t *dst, ptrdiff_t dststride,
                              uint8_t *src, ptrdiff_t srcstride,
                              int height, int mx, int my);
-void ff_put_vp8_pixels16_mmx(uint8_t *dst, ptrdiff_t dststride,
-                             uint8_t *src, ptrdiff_t srcstride,
-                             int height, int mx, int my);
 void ff_put_vp8_pixels16_sse(uint8_t *dst, ptrdiff_t dststride,
                              uint8_t *src, ptrdiff_t srcstride,
                              int height, int mx, int my);
@@ -140,19 +137,6 @@ static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
         dst + 4, dststride, src + 4, srcstride, height, mx, my); \
 }
 
-#if ARCH_X86_32
-TAP_W8 (mmxext, epel, h4)
-TAP_W8 (mmxext, epel, h6)
-TAP_W16(mmxext, epel, h6)
-TAP_W8 (mmxext, epel, v4)
-TAP_W8 (mmxext, epel, v6)
-TAP_W16(mmxext, epel, v6)
-TAP_W8 (mmxext, bilinear, h)
-TAP_W16(mmxext, bilinear, h)
-TAP_W8 (mmxext, bilinear, v)
-TAP_W16(mmxext, bilinear, v)
-#endif
-
 TAP_W16(sse2,  epel, h6)
 TAP_W16(sse2,  epel, v6)
 TAP_W16(sse2,  bilinear, h)
@@ -177,16 +161,8 @@ static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT
         dst, dststride, tmpptr, SIZE,      height,               mx, my); \
 }
 
-#if ARCH_X86_32
-#define HVTAPMMX(x, y) \
-HVTAP(mmxext, 8, x, y,  4,  8) \
-HVTAP(mmxext, 8, x, y,  8, 16)
-
-HVTAP(mmxext, 8, 6, 6, 16, 16)
-#else
 #define HVTAPMMX(x, y) \
 HVTAP(mmxext, 8, x, y,  4,  8)
-#endif
 
 HVTAPMMX(4, 4)
 HVTAPMMX(4, 6)
@@ -221,31 +197,21 @@ static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \
 }
 
 HVBILIN(mmxext,  8,  4,  8)
-#if ARCH_X86_32
-HVBILIN(mmxext,  8,  8, 16)
-HVBILIN(mmxext,  8, 16, 16)
-#endif
 HVBILIN(sse2,  8,  8, 16)
 HVBILIN(sse2,  8, 16, 16)
 HVBILIN(ssse3, 8,  4,  8)
 HVBILIN(ssse3, 8,  8, 16)
 HVBILIN(ssse3, 8, 16, 16)
 
-void ff_vp8_idct_dc_add_mmx(uint8_t *dst, int16_t block[16],
-                            ptrdiff_t stride);
 void ff_vp8_idct_dc_add_sse2(uint8_t *dst, int16_t block[16],
                              ptrdiff_t stride);
 void ff_vp8_idct_dc_add_sse4(uint8_t *dst, int16_t block[16],
                              ptrdiff_t stride);
-void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, int16_t block[4][16],
-                               ptrdiff_t stride);
 void ff_vp8_idct_dc_add4y_sse2(uint8_t *dst, int16_t block[4][16],
                                ptrdiff_t stride);
 void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, int16_t block[2][16],
                                ptrdiff_t stride);
-void ff_vp8_luma_dc_wht_mmx(int16_t block[4][4][16], int16_t dc[16]);
 void ff_vp8_luma_dc_wht_sse(int16_t block[4][4][16], int16_t dc[16]);
-void ff_vp8_idct_add_mmx(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
 void ff_vp8_idct_add_sse(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
 
 #define DECLARE_LOOP_FILTER(NAME)                                       \
@@ -284,8 +250,6 @@ void ff_vp8_h_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU,             \
                                              ptrdiff_t s,               \
                                              int e, int i, int hvt);
 
-DECLARE_LOOP_FILTER(mmx)
-DECLARE_LOOP_FILTER(mmxext)
 DECLARE_LOOP_FILTER(sse2)
 DECLARE_LOOP_FILTER(ssse3)
 DECLARE_LOOP_FILTER(sse4)
@@ -322,10 +286,6 @@ av_cold void ff_vp78dsp_init_x86(VP8DSPContext *c)
     int cpu_flags = av_get_cpu_flags();
 
     if (EXTERNAL_MMX(cpu_flags)) {
-#if ARCH_X86_32
-        c->put_vp8_epel_pixels_tab[0][0][0]     =
-        c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
-#endif
         c->put_vp8_epel_pixels_tab[1][0][0]     =
         c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
     }
@@ -335,12 +295,6 @@ av_cold void ff_vp78dsp_init_x86(VP8DSPContext *c)
     if (EXTERNAL_MMXEXT(cpu_flags)) {
         VP8_MC_FUNC(2, 4, mmxext);
         VP8_BILINEAR_MC_FUNC(2, 4, mmxext);
-#if ARCH_X86_32
-        VP8_LUMA_MC_FUNC(0, 16, mmxext);
-        VP8_MC_FUNC(1, 8, mmxext);
-        VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
-        VP8_BILINEAR_MC_FUNC(1,  8, mmxext);
-#endif
     }
 
     if (EXTERNAL_SSE(cpu_flags)) {
@@ -373,44 +327,6 @@ av_cold void ff_vp8dsp_init_x86(VP8DSPContext *c)
 
     if (EXTERNAL_MMX(cpu_flags)) {
         c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmx;
-#if ARCH_X86_32
-        c->vp8_idct_dc_add    = ff_vp8_idct_dc_add_mmx;
-        c->vp8_idct_dc_add4y  = ff_vp8_idct_dc_add4y_mmx;
-        c->vp8_idct_add       = ff_vp8_idct_add_mmx;
-        c->vp8_luma_dc_wht    = ff_vp8_luma_dc_wht_mmx;
-
-        c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx;
-        c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx;
-
-        c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx;
-        c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx;
-        c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmx;
-        c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmx;
-
-        c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_mmx;
-        c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_mmx;
-        c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_mmx;
-        c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_mmx;
-#endif
-    }
-
-    /* note that 4-tap width=16 functions are missing because w=16
-     * is only used for luma, and luma is always a copy or sixtap. */
-    if (EXTERNAL_MMXEXT(cpu_flags)) {
-#if ARCH_X86_32
-        c->vp8_v_loop_filter_simple   = ff_vp8_v_loop_filter_simple_mmxext;
-        c->vp8_h_loop_filter_simple   = ff_vp8_h_loop_filter_simple_mmxext;
-
-        c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext;
-        c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext;
-        c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmxext;
-        c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmxext;
-
-        c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_mmxext;
-        c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_mmxext;
-        c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_mmxext;
-        c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_mmxext;
-#endif
     }
 
     if (EXTERNAL_SSE(cpu_flags)) {
diff --git a/libavcodec/x86/vp8dsp_loopfilter.asm b/libavcodec/x86/vp8dsp_loopfilter.asm
index caeb405267..ef397efd3e 100644
--- a/libavcodec/x86/vp8dsp_loopfilter.asm
+++ b/libavcodec/x86/vp8dsp_loopfilter.asm
@@ -46,30 +46,6 @@ SECTION .text
 ; void ff_vp8_h/v_loop_filter_simple_<opt>(uint8_t *dst, ptrdiff_t stride, int flim);
 ;-----------------------------------------------------------------------------
 
-; macro called with 7 mm register indexes as argument, and 4 regular registers
-;
-; first 4 mm registers will carry the transposed pixel data
-; the other three are scratchspace (one would be sufficient, but this allows
-; for more spreading/pipelining and thus faster execution on OOE CPUs)
-;
-; first two regular registers are buf+4*stride and buf+5*stride
-; third is -stride, fourth is +stride
-%macro READ_8x4_INTERLEAVED 11
-    ; interleave 8 (A-H) rows of 4 pixels each
-    movd          m%1, [%8+%10*4]   ; A0-3
-    movd          m%5, [%9+%10*4]   ; B0-3
-    movd          m%2, [%8+%10*2]   ; C0-3
-    movd          m%6, [%8+%10]     ; D0-3
-    movd          m%3, [%8]         ; E0-3
-    movd          m%7, [%9]         ; F0-3
-    movd          m%4, [%9+%11]     ; G0-3
-    punpcklbw     m%1, m%5          ; A/B interleaved
-    movd          m%5, [%9+%11*2]   ; H0-3
-    punpcklbw     m%2, m%6          ; C/D interleaved
-    punpcklbw     m%3, m%7          ; E/F interleaved
-    punpcklbw     m%4, m%5          ; G/H interleaved
-%endmacro
-
 ; macro called with 7 mm register indexes as argument, and 5 regular registers
 ; first 11 mean the same as READ_8x4_TRANSPOSED above
 ; fifth regular register is scratchspace to reach the bottom 8 rows, it
@@ -112,26 +88,6 @@ SECTION .text
     punpcklbw    m%4, m%5          ; G/H/O/P interleaved
 %endmacro
 
-; write 4 mm registers of 2 dwords each
-; first four arguments are mm register indexes containing source data
-; last four are registers containing buf+4*stride, buf+5*stride,
-; -stride and +stride
-%macro WRITE_4x2D 8
-    ; write out (2 dwords per register)
-    movd    [%5+%7*4], m%1
-    movd    [%5+%7*2], m%2
-    movd         [%5], m%3
-    movd      [%6+%8], m%4
-    punpckhdq     m%1, m%1
-    punpckhdq     m%2, m%2
-    punpckhdq     m%3, m%3
-    punpckhdq     m%4, m%4
-    movd    [%6+%7*4], m%1
-    movd      [%5+%7], m%2
-    movd         [%6], m%3
-    movd    [%6+%8*2], m%4
-%endmacro
-
 ; write 4 xmm registers of 4 dwords each
 ; arguments same as WRITE_2x4D, but with an extra register, so that the 5 regular
 ; registers contain buf+4*stride, buf+5*stride, buf+12*stride, -stride and +stride
@@ -192,42 +148,6 @@ SECTION .text
     movd    [%7+%9*2], m%4
 %endmacro
 
-; write 4 or 8 words in the mmx/xmm registers as 8 lines
-; 1 and 2 are the registers to write, this can be the same (for SSE2)
-; for pre-SSE4:
-; 3 is a general-purpose register that we will clobber
-; for SSE4:
-; 3 is a pointer to the destination's 5th line
-; 4 is a pointer to the destination's 4th line
-; 5/6 is -stride and +stride
-%macro WRITE_2x4W 6
-    movd            %3d, %1
-    punpckhdq        %1, %1
-    mov       [%4+%5*4], %3w
-    shr              %3, 16
-    add              %4, %6
-    mov       [%4+%5*4], %3w
-
-    movd            %3d, %1
-    add              %4, %5
-    mov       [%4+%5*2], %3w
-    shr              %3, 16
-    mov       [%4+%5  ], %3w
-
-    movd            %3d, %2
-    punpckhdq        %2, %2
-    mov       [%4     ], %3w
-    shr              %3, 16
-    mov       [%4+%6  ], %3w
-
-    movd            %3d, %2
-    add              %4, %6
-    mov       [%4+%6  ], %3w
-    shr              %3, 16
-    mov       [%4+%6*2], %3w
-    add              %4, %5
-%endmacro
-
 %macro WRITE_8W 5
 %if cpuflag(sse4)
     pextrw    [%3+%4*4], %1, 0
@@ -269,29 +189,19 @@ SECTION .text
 
 %macro SIMPLE_LOOPFILTER 2
 cglobal vp8_%1_loop_filter_simple, 3, %2, 8, dst, stride, flim, cntr
-%if mmsize == 8 ; mmx/mmxext
-    mov         cntrq, 2
-%endif
 %if cpuflag(ssse3)
     pxor           m0, m0
 %endif
     SPLATB_REG     m7, flim, m0     ; splat "flim" into register
 
     ; set up indexes to address 4 rows
-%if mmsize == 8
-    DEFINE_ARGS dst1, mstride, stride, cntr, dst2
-%else
     DEFINE_ARGS dst1, mstride, stride, dst3, dst2
-%endif
     mov       strideq, mstrideq
     neg      mstrideq
 %ifidn %1, h
     lea         dst1q, [dst1q+4*strideq-2]
 %endif
 
-%if mmsize == 8 ; mmx / mmxext
-.next8px:
-%endif
 %ifidn %1, v
     ; read 4 half/full rows of pixels
     mova           m0, [dst1q+mstrideq*2]    ; p1
@@ -301,11 +211,7 @@ cglobal vp8_%1_loop_filter_simple, 3, %2, 8, dst, stride, flim, cntr
 %else ; h
     lea         dst2q, [dst1q+ strideq]
 
-%if mmsize == 8 ; mmx/mmxext
-    READ_8x4_INTERLEAVED  0, 1, 2, 3, 4, 5, 6, dst1q, dst2q, mstrideq, strideq
-%else ; sse2
     READ_16x4_INTERLEAVED 0, 1, 2, 3, 4, 5, 6, dst1q, dst2q, mstrideq, strideq, dst3q
-%endif
     TRANSPOSE4x4W         0, 1, 2, 3, 4
 %endif
 
@@ -380,7 +286,6 @@ cglobal vp8_%1_loop_filter_simple, 3, %2, 8, dst, stride, flim, cntr
     inc        dst1q
     SBUTTERFLY    bw, 6, 4, 0
 
-%if mmsize == 16 ; sse2
 %if cpuflag(sse4)
     inc         dst2q
 %endif
@@ -390,35 +295,11 @@ cglobal vp8_%1_loop_filter_simple, 3, %2, 8, dst, stride, flim, cntr
     inc         dst3q
 %endif
     WRITE_8W       m4, dst3q, dst2q, mstrideq, strideq
-%else ; mmx/mmxext
-    WRITE_2x4W     m6, m4, dst2q, dst1q, mstrideq, strideq
-%endif
 %endif
 
-%if mmsize == 8 ; mmx/mmxext
-    ; next 8 pixels
-%ifidn %1, v
-    add         dst1q, 8            ; advance 8 cols = pixels
-%else ; h
-    lea         dst1q, [dst1q+strideq*8-1]  ; advance 8 rows = lines
-%endif
-    dec         cntrq
-    jg .next8px
-    REP_RET
-%else ; sse2
     RET
-%endif
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-SIMPLE_LOOPFILTER v, 4
-SIMPLE_LOOPFILTER h, 5
-INIT_MMX mmxext
-SIMPLE_LOOPFILTER v, 4
-SIMPLE_LOOPFILTER h, 5
-%endif
-
 INIT_XMM sse2
 SIMPLE_LOOPFILTER v, 3
 SIMPLE_LOOPFILTER h, 5
@@ -485,9 +366,6 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
 
 %if %2 == 8 ; chroma
     DEFINE_ARGS dst1, dst8, mstride, stride, dst2
-%elif mmsize == 8
-    DEFINE_ARGS dst1, mstride, stride, dst2, cntr
-    mov           cntrq, 2
 %else
     DEFINE_ARGS dst1, mstride, stride, dst2, dst8
 %endif
@@ -500,9 +378,6 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
 %endif
 %endif
 
-%if mmsize == 8
-.next8px:
-%endif
     ; read
     lea           dst2q, [dst1q+strideq]
 %ifidn %1, v
@@ -527,33 +402,7 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     movhps           m7, [dst8q+ strideq*2]
     add           dst8q, mstrideq
 %endif
-%elif mmsize == 8 ; mmx/mmxext (h)
-    ; read 8 rows of 8px each
-    movu             m0, [dst1q+mstrideq*4]
-    movu             m1, [dst2q+mstrideq*4]
-    movu             m2, [dst1q+mstrideq*2]
-    movu             m3, [dst1q+mstrideq  ]
-    movu             m4, [dst1q]
-    movu             m5, [dst2q]
-    movu             m6, [dst2q+ strideq  ]
-
-    ; 8x8 transpose
-    TRANSPOSE4x4B     0, 1, 2, 3, 7
-    mova     m_q0backup, m1
-    movu             m7, [dst2q+ strideq*2]
-    TRANSPOSE4x4B     4, 5, 6, 7, 1
-    SBUTTERFLY       dq, 0, 4, 1     ; p3/p2
-    SBUTTERFLY       dq, 2, 6, 1     ; q0/q1
-    SBUTTERFLY       dq, 3, 7, 1     ; q2/q3
-    mova             m1, m_q0backup
-    mova     m_q0backup, m2          ; store q0
-    SBUTTERFLY       dq, 1, 5, 2     ; p1/p0
-    mova     m_p0backup, m5          ; store p0
-    SWAP              1, 4
-    SWAP              2, 4
-    SWAP              6, 3
-    SWAP              5, 3
-%else ; sse2 (h)
+%else ; h
 %if %2 == 16
     lea           dst8q, [dst1q+ strideq*8]
 %endif
@@ -641,25 +490,9 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     psubusb          m6, m5          ; q2-q1
     por              m6, m4          ; abs(q2-q1)
 
-%if notcpuflag(mmxext)
-    mova             m4, m_flimI
-    pxor             m3, m3
-    psubusb          m0, m4
-    psubusb          m1, m4
-    psubusb          m7, m4
-    psubusb          m6, m4
-    pcmpeqb          m0, m3          ; abs(p3-p2) <= I
-    pcmpeqb          m1, m3          ; abs(p2-p1) <= I
-    pcmpeqb          m7, m3          ; abs(q3-q2) <= I
-    pcmpeqb          m6, m3          ; abs(q2-q1) <= I
-    pand             m0, m1
-    pand             m7, m6
-    pand             m0, m7
-%else ; mmxext/sse2
     pmaxub           m0, m1
     pmaxub           m6, m7
     pmaxub           m0, m6
-%endif
 
     ; normal_limit and high_edge_variance for p1-p0, q1-q0
     SWAP              7, 3           ; now m7 is zero
@@ -681,18 +514,8 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     psubusb          m1, m3          ; p1-p0
     psubusb          m6, m2          ; p0-p1
     por              m1, m6          ; abs(p1-p0)
-%if notcpuflag(mmxext)
-    mova             m6, m1
-    psubusb          m1, m4
-    psubusb          m6, m_hevthr
-    pcmpeqb          m1, m7          ; abs(p1-p0) <= I
-    pcmpeqb          m6, m7          ; abs(p1-p0) <= hev_thresh
-    pand             m0, m1
-    mova      m_maskres, m6
-%else ; mmxext/sse2
     pmaxub           m0, m1          ; max_I
     SWAP              1, 4           ; max_hev_thresh
-%endif
 
     SWAP              6, 4           ; now m6 is I
 %ifidn %1, v
@@ -712,17 +535,6 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     psubusb          m1, m5          ; q0-q1
     psubusb          m7, m4          ; q1-q0
     por              m1, m7          ; abs(q1-q0)
-%if notcpuflag(mmxext)
-    mova             m7, m1
-    psubusb          m1, m6
-    psubusb          m7, m_hevthr
-    pxor             m6, m6
-    pcmpeqb          m1, m6          ; abs(q1-q0) <= I
-    pcmpeqb          m7, m6          ; abs(q1-q0) <= hev_thresh
-    mova             m6, m_maskres
-    pand             m0, m1          ; abs([pq][321]-[pq][210]) <= I
-    pand             m6, m7
-%else ; mmxext/sse2
     pxor             m7, m7
     pmaxub           m0, m1
     pmaxub           m6, m1
@@ -730,7 +542,6 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     psubusb          m6, m_hevthr
     pcmpeqb          m0, m7          ; max(abs(..)) <= I
     pcmpeqb          m6, m7          ; !(max(abs..) > thresh)
-%endif
 %ifdef m12
     SWAP              6, 12
 %else
@@ -820,25 +631,12 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
 %else
     mova             m6, m_maskres
 %endif
-%if notcpuflag(mmxext)
-    mova             m7, [pb_1]
-%else ; mmxext/sse2
     pxor             m7, m7
-%endif
     pand             m0, m6
     pand             m1, m6
-%if notcpuflag(mmxext)
-    paddusb          m0, m7
-    pand             m1, [pb_FE]
-    pandn            m7, m0
-    psrlq            m1, 1
-    psrlq            m7, 1
-    SWAP              0, 7
-%else ; mmxext/sse2
     psubusb          m1, [pb_1]
     pavgb            m0, m7          ; a
     pavgb            m1, m7          ; -a
-%endif
     psubusb          m5, m0
     psubusb          m2, m1
     paddusb          m5, m1          ; q1-a
@@ -863,51 +661,13 @@ cglobal vp8_%1_loop_filter16y_inner, 5, 5, 13, stack_size, dst, stride, flimE, f
     ; 4x8/16 transpose
     TRANSPOSE4x4B     2, 3, 4, 5, 6
 
-%if mmsize == 8 ; mmx/mmxext (h)
-    WRITE_4x2D        2, 3, 4, 5, dst1q, dst2q, mstrideq, strideq
-%else ; sse2 (h)
     lea           dst8q, [dst8q+mstrideq  +2]
     WRITE_4x4D        2, 3, 4, 5, dst1q, dst2q, dst8q, mstrideq, strideq, %2
 %endif
-%endif
 
-%if mmsize == 8
-%if %2 == 8 ; chroma
-%ifidn %1, h
-    sub           dst1q, 2
-%endif
-    cmp           dst1q, dst8q
-    mov           dst1q, dst8q
-    jnz .next8px
-%else
-%ifidn %1, h
-    lea           dst1q, [dst1q+ strideq*8-2]
-%else ; v
-    add           dst1q, 8
-%endif
-    dec           cntrq
-    jg .next8px
-%endif
-    REP_RET
-%else ; mmsize == 16
     RET
-%endif
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-INNER_LOOPFILTER v, 16
-INNER_LOOPFILTER h, 16
-INNER_LOOPFILTER v,  8
-INNER_LOOPFILTER h,  8
-
-INIT_MMX mmxext
-INNER_LOOPFILTER v, 16
-INNER_LOOPFILTER h, 16
-INNER_LOOPFILTER v,  8
-INNER_LOOPFILTER h,  8
-%endif
-
 INIT_XMM sse2
 INNER_LOOPFILTER v, 16
 INNER_LOOPFILTER h, 16
@@ -992,9 +752,6 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
 
 %if %2 == 8 ; chroma
     DEFINE_ARGS dst1, dst8, mstride, stride, dst2
-%elif mmsize == 8
-    DEFINE_ARGS dst1, mstride, stride, dst2, cntr
-    mov           cntrq, 2
 %else
     DEFINE_ARGS dst1, mstride, stride, dst2, dst8
 %endif
@@ -1007,9 +764,6 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
 %endif
 %endif
 
-%if mmsize == 8
-.next8px:
-%endif
     ; read
     lea           dst2q, [dst1q+ strideq  ]
 %ifidn %1, v
@@ -1034,33 +788,7 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     movhps           m7, [dst8q+ strideq*2]
     add           dst8q, mstrideq
 %endif
-%elif mmsize == 8 ; mmx/mmxext (h)
-    ; read 8 rows of 8px each
-    movu             m0, [dst1q+mstrideq*4]
-    movu             m1, [dst2q+mstrideq*4]
-    movu             m2, [dst1q+mstrideq*2]
-    movu             m3, [dst1q+mstrideq  ]
-    movu             m4, [dst1q]
-    movu             m5, [dst2q]
-    movu             m6, [dst2q+ strideq  ]
-
-    ; 8x8 transpose
-    TRANSPOSE4x4B     0, 1, 2, 3, 7
-    mova     m_q0backup, m1
-    movu             m7, [dst2q+ strideq*2]
-    TRANSPOSE4x4B     4, 5, 6, 7, 1
-    SBUTTERFLY       dq, 0, 4, 1     ; p3/p2
-    SBUTTERFLY       dq, 2, 6, 1     ; q0/q1
-    SBUTTERFLY       dq, 3, 7, 1     ; q2/q3
-    mova             m1, m_q0backup
-    mova     m_q0backup, m2          ; store q0
-    SBUTTERFLY       dq, 1, 5, 2     ; p1/p0
-    mova     m_p0backup, m5          ; store p0
-    SWAP              1, 4
-    SWAP              2, 4
-    SWAP              6, 3
-    SWAP              5, 3
-%else ; sse2 (h)
+%else ; h
 %if %2 == 16
     lea           dst8q, [dst1q+ strideq*8  ]
 %endif
@@ -1150,25 +878,9 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     psubusb          m6, m5          ; q2-q1
     por              m6, m4          ; abs(q2-q1)
 
-%if notcpuflag(mmxext)
-    mova             m4, m_flimI
-    pxor             m3, m3
-    psubusb          m0, m4
-    psubusb          m1, m4
-    psubusb          m7, m4
-    psubusb          m6, m4
-    pcmpeqb          m0, m3          ; abs(p3-p2) <= I
-    pcmpeqb          m1, m3          ; abs(p2-p1) <= I
-    pcmpeqb          m7, m3          ; abs(q3-q2) <= I
-    pcmpeqb          m6, m3          ; abs(q2-q1) <= I
-    pand             m0, m1
-    pand             m7, m6
-    pand             m0, m7
-%else ; mmxext/sse2
     pmaxub           m0, m1
     pmaxub           m6, m7
     pmaxub           m0, m6
-%endif
 
     ; normal_limit and high_edge_variance for p1-p0, q1-q0
     SWAP              7, 3           ; now m7 is zero
@@ -1190,18 +902,8 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     psubusb          m1, m3          ; p1-p0
     psubusb          m6, m2          ; p0-p1
     por              m1, m6          ; abs(p1-p0)
-%if notcpuflag(mmxext)
-    mova             m6, m1
-    psubusb          m1, m4
-    psubusb          m6, m_hevthr
-    pcmpeqb          m1, m7          ; abs(p1-p0) <= I
-    pcmpeqb          m6, m7          ; abs(p1-p0) <= hev_thresh
-    pand             m0, m1
-    mova      m_maskres, m6
-%else ; mmxext/sse2
     pmaxub           m0, m1          ; max_I
     SWAP              1, 4           ; max_hev_thresh
-%endif
 
     SWAP              6, 4           ; now m6 is I
 %ifidn %1, v
@@ -1221,17 +923,6 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     psubusb          m1, m5          ; q0-q1
     psubusb          m7, m4          ; q1-q0
     por              m1, m7          ; abs(q1-q0)
-%if notcpuflag(mmxext)
-    mova             m7, m1
-    psubusb          m1, m6
-    psubusb          m7, m_hevthr
-    pxor             m6, m6
-    pcmpeqb          m1, m6          ; abs(q1-q0) <= I
-    pcmpeqb          m7, m6          ; abs(q1-q0) <= hev_thresh
-    mova             m6, m_maskres
-    pand             m0, m1          ; abs([pq][321]-[pq][210]) <= I
-    pand             m6, m7
-%else ; mmxext/sse2
     pxor             m7, m7
     pmaxub           m0, m1
     pmaxub           m6, m1
@@ -1239,7 +930,6 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     psubusb          m6, m_hevthr
     pcmpeqb          m0, m7          ; max(abs(..)) <= I
     pcmpeqb          m6, m7          ; !(max(abs..) > thresh)
-%endif
 %ifdef m12
     SWAP              6, 12
 %else
@@ -1510,11 +1200,6 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
     TRANSPOSE4x4B    1, 2, 3, 4, 0
     SBUTTERFLY      bw, 5, 6, 0
 
-%if mmsize == 8 ; mmx/mmxext (h)
-    WRITE_4x2D       1, 2, 3, 4, dst1q, dst2q, mstrideq, strideq
-    add          dst1q, 4
-    WRITE_2x4W      m5, m6, dst2q, dst1q, mstrideq, strideq
-%else ; sse2 (h)
     lea          dst8q, [dst8q+mstrideq+1]
     WRITE_4x4D       1, 2, 3, 4, dst1q, dst2q, dst8q, mstrideq, strideq, %2
     lea          dst1q, [dst2q+mstrideq+4]
@@ -1528,45 +1213,10 @@ cglobal vp8_%1_loop_filter16y_mbedge, 5, 5, 15, stack_size, dst1, stride, flimE,
 %endif
     WRITE_8W        m6, dst2q, dst8q, mstrideq, strideq
 %endif
-%endif
 
-%if mmsize == 8
-%if %2 == 8 ; chroma
-%ifidn %1, h
-    sub          dst1q, 5
-%endif
-    cmp          dst1q, dst8q
-    mov          dst1q, dst8q
-    jnz .next8px
-%else
-%ifidn %1, h
-    lea          dst1q, [dst1q+ strideq*8-5]
-%else ; v
-    add          dst1q, 8
-%endif
-    dec          cntrq
-    jg .next8px
-%endif
-    REP_RET
-%else ; mmsize == 16
     RET
-%endif
 %endmacro
 
-%if ARCH_X86_32
-INIT_MMX mmx
-MBEDGE_LOOPFILTER v, 16
-MBEDGE_LOOPFILTER h, 16
-MBEDGE_LOOPFILTER v,  8
-MBEDGE_LOOPFILTER h,  8
-
-INIT_MMX mmxext
-MBEDGE_LOOPFILTER v, 16
-MBEDGE_LOOPFILTER h, 16
-MBEDGE_LOOPFILTER v,  8
-MBEDGE_LOOPFILTER h,  8
-%endif
-
 INIT_XMM sse2
 MBEDGE_LOOPFILTER v, 16
 MBEDGE_LOOPFILTER h, 16